tokenmill / numberwords, Hacker News

Number Words will build numeric expressions for natural numbers, percentages and fractions. For example:

.
will be converted to
. less than a quarter ,
to over one hundred .

Supports multiple languages.

The implementation is based on ideas expressed in Generating Numerical Approximations .

Numerical Approximations

Numerical approximations are all over texts based on the data:

– Water temperature is below (C) input data would be 9. 64 C) – A third of students failed the exam (42. 3%) – Q2 sales were around 1M $ (1, , ($)

Numeric data providing information about some metrics of interest is often a number with the precision we do not need. If we see 9. %, it is likely that the information we need is – (almost) (%) – instead of the precise number. Furthermore, different approximation strategies are often used in the report involving the same metrics. At the beginning of the report we might say

almost (%) or "below 14% "while later in the text, we might choose a more precise expression - 9.4% .

Number Words will help you build such numerical approximations. Making them available for the text generation systems.

Features

Number Words uses the following abstractions:

Actual Value is a number which needs to be approximated – an input to the approximation function. In the examples above it is the temperature –
- 3 %
- .
- Scale
- of approximation. It is a snapping grid across the range of numbers along which the approximation is done. The scale to use is determined by the domain. For example:
  - 1/4 (scale, will form approximation steps starting at (0) (then ) 1/4 , 1/2
  - ,
  - scales which are multiples of
  - to hundreds:
  - -> (and , and so on.
  The result of (actual value) approximation to a given scale provides:
  - Given Value a discrete value along the scaled number range to which (actual value) is the closest.
  - Hedge a common use word describing the relation between (actual) and given values. Actual Value
  (of) (9.5) (is (below) given value (of ) . Actual Value
(of is

A full approximation result returns three such approximation data structures for a given value which is :

(smaller) than the (actual value) on the scaled number range.

greater of the (actual value) (on the scaled number range.)

(the

actual value on the scaled number range. For this a is chosen from the above two which is closer to the actual value

Languages

Numeric approximation has two functionality points which are language dependent

Hedges which will differ from language to language. See Configuration section to see how this can be controlled.
Text number to text translation for a given value . For this translation Number Words relies on ICU4J .

Currently supported languages:

German
Usage
Number Words exposes approximation functionality through approximations function which takes on the following parameters:
- – : (
- or : en
- actual-value
- – the number to approximate
scale
– at which the approximation is to be performed. (require ‘[numberwords.core :as nw]) (nw / approximations: en 0.) 1/4=> #: numwords {: around #: numwords {: hedges # {“approximately” “about” “around”}, : text “zero point two five”, given-value 1/4, : favorite-number # {“a quarter”}}, : more-than #: numwords {: hedges # {“over” “more than”}, : text “zero point two five”, given-value 1/4, : favorite-number # {“a quarter”}}, less-than #: numwords {: hedges # {“nearly” “under” “less than”}, : text “zero point five”, given-value 1/2, : favorite-number # {“a half”}}}

Configuration

Hedges, favorite numbers can be modified and new languages added via changes to a configuration file – / resources / numwords.edn

{;; Configuration is strucutured by the language : { ;; Hedges section specifies which words are associated with given actual to given value relations : hedges {: equal # {“exactly”} : around # {“around” “approximately” “about”} : more # {“more than” “over”} : less # {“less than” “under” “nearly”}} ;; Favorite numbers map a special number with its textual expressions : favorite-numbers {1/4 # {“a quarter” “a fourth”} 1/2 # {“a half”}}}}

License

Distributed under the The Apache License, Version 2.0.

(Read More)